Intrusion Detection Based on Adaptive Immune Classifier

The distribution properties of the normally data and anomaly data in the network connectivity features have huge differences ; therefore, there exist the low rate of detection and false positive rate problem for the traditional classifier which is applied to the network intrusion detection. An adaptive classifier based on the artificial immune cluster is presented. The new classifier adopts multi -granularities idea and it effectively eliminates the inconsistency between the classification algorithm and the clustering algorithm. Through the classification of the data sets in real variety of network intrusion data sets, experimental results show that the classifier has high detection rate and low false positive rate; it has better classification performance and generalization ability than RBF and BP classifiers.     

Face recognition method with local matching

From human cognition, a face recognition method with local matching based on statistical learning is proposed. The image is divided into several subimages and each subimage is considered as a weak classifier. The Adaboost learning algorithm is used to train the weak classifiers and construct a strong classifier. As a result, all subimages are effectively combined together to explore the best discriminating power and improve the classification accuracy. Compared with the holistic matching methods, the local matching method is robust to variations in illumination, expression, and pose, etc. The experimental results show that the proposed method can improve the face recognition accuracy and is robust to variations in illumination and expression.     

基于Adaboost算法的输电线路舞动预警方法

输电线路舞动是目前尚未被全面准确认识的世界性难题,已严重威胁输电系统的安全稳定运行。文章分析影响舞动的外界气象环境因素,并在此基础上提出一种基于Adaboost集成学习算法的输电线舞动预警方法。采用基于Gini指标的决策桩作为弱分类器,通过对多个弱分类器的训练及加权求和,输出舞动预测结果及其置信度,可为电网运维人员提供决策支撑。最后,使用历史数据进行验证性实验,结果证明了所提方法的有效性。     

An edge detection method based on a good point set genetic algorithm

In order to improve the convergence rate of genetic algorithms based on edge detection, a novel edge detection method based on a good point set genetic algorithm (GGA) was proposed. The proposed method designed the crossover operation with the theory of good point set in which the progeny inherits the common genes of the parents which represent its family so as to improve the convergence rate of the genetic algorithm. Furthermore, before the algorithm was used for edge detection, the feature space of the image grey level was transformed into the feature space of the fuzzy entropy. Dissimilarity enhancement processing next was applied to the image by using a fuzzy entropy theory to filter the non edge pixels so as to reduce the scale of the solution domain. This approach offered another efficient way to improve the convergence rate. Experimental results show the proposed algorithm performs very well in terms of convergence rate. The detected edge image is well localized, thin, and robustly resistant to noise.     

Two-mode clustering of genotype by trait and genotype by environment data

In this paper, we demonstrate the use of two-mode clustering for genotype by trait and genotype by environment data. In contrast to two separate (one mode) clusterings on genotypes or traits/environments, two-mode clustering simultaneously produces homogeneous groups of genotypes and traits/environments. For two-mode clustering, we first scan all two-mode cluster solutions with all possible numbers of clusters using k-means. After deciding on the final numbers of clusters, we continue with a two-mode clustering algorithm based on a genetic algorithm. This ensures optimal solutions even for large data sets. We discuss the application of two-mode clustering to multiple trait data stemming from genomic research on tomatoes as well as an application to multi-environment data on barley.     

Fault diagnosis of rotating machinery based on fuzzy clusteringoptimized by chaos embedded particle swarm optimization

A method of weighted fuzzy clustering optimized by chaos embedded particle swarm algorithm(CPSO) is put forward and applied in vibration fault diagnosis of rotating machinery. In the method, CPSO is used to displace the traditional stochastic-gradient algorithm to optimize parameters of weighted fuzzy C-means (WFCM). The best clustering num and clustering centers are automatically attained according to clustering validity function. The experimental results show that the method effectively increases the convergence velocity and precision of WFCM and so does the correctness rate of fault diagnosis for rotating machinery.     

Bottom-up Clustering Algorithm of High Dimension

Clustering algorithm based on wavelet transform is efficient, and which can detect clusters of arbitrary shape. It is insensitive to the outliers and the order of input data. However, efficiency of the algorithm would be degraded, and computation complexity of the algorithm would be considerable with increase of clustering dimensions. A bottom-u Pmethod is put forward to make the original algorithm fit to clustering in high dimension, and the scalability of the improved algorithm is enhanced by parallelization. The experiment demonstrates that the improved algorithm has no impact on quality of clustering and has a good efficient in high dimension clustering and in decrease of comnutation comnlexity.     

Algorithm for Mining Sequential Pattern

Mining sequential pattern is an important topic in the data mining research. In this paper, on the basis of recording the Ctid scheme of the set in every frequent set, the authors propose an algorithm named ISP for mining sequential pattern. In the algorithm the items and the sequence are discussed respectively, and the time join method is used to introduce the candidate sets, so the frequent sets can be gotten. The ISP algorithm takes full use of the existing and updated Ctid scheme, therefore the efficiency of the process is increased besides guaranteeing the validity of the algorithm. Comparing with the algorithm named IMSP,more efficient rules are obtained.     

Cotton Image Segmentation Algorithm Based on Fusion Method of Markov Random Field and Quantum Particle Swarm Cluster

[Objective] The aim of this study was to improve the cotton image segmentation accuracy in a picking robot image processing system. [Method] An image segmentation algorithm based on a fusion method of Markov random field and quantum particle swarm optimization clustering was proposed. The process of the proposed algorithm is as follows: first, transform the RGB (red, green, blue) images into grayscale; second, use it to segment these images; finally, the threshold of the connected area is set on the basis of the segmented image to obtain the target area. Then, the cotton front image and the cotton side image are selected from the images collected from different angles. The segmentation experiment was carried out by using this algorithm, and compared with the Otsu algorithm, the fuzzy C-means algorithm, the quantum particle swarm image segmentation algorithm and the Markov random field image segmentation algorithm. [Result] The results showed that the segmentation accuracy and peak signal to noise ratio of the proposed algorithm were 98.94% and 77.48 dB. When compared with the Otsu algorithm, fuzzy C-means algorithm, quantum particle swarm optimization algorithm and Markov random field algorithm, the average segmentation accuracy and peak signal to noise ratio of the proposed algorithm increased by 2.47%–4.56%, and 9.81–13.11 dB, respectively. [Conclusion] The proposed algorithm had higher segmentation accuracy and higher peak signal to noise ratio than the other algorithms tested.     

Thermal process model identification based on radial basisfunction neural networks

In order to accurately reflect the dynamic behavior and realize the whole optimal control of the thermal process, a novel modeling method of the RBF NN (Radial Basis Function Neural Networks) model is proposed to build nonlinear model. This method is based on entropy clustering and competitive learning algorithm, combined with nonlinear autoregressive moving average (NARMA) model to identify the RBF NN stucture, and the power vector is gotten by the least square algorithm. Two simulation experiments show that the proposed method of the identification based on NARMA model and RBF NN can accurately describe the non linearity of the process and has less hidden nodes.     

Information Fusion Algorithm Based Evidence Combination

This paper applies the evidence combination theory to fuse the information of multi-neural network classifier. In order to make each classifier approach the ideal state, the heredity algorithm is applied to train it. The different capacity of each classifier is caused by different classified feature. Input feature can't be identified by one classifier and may be identified by another, Model identification can be performed by multi-classifier, output result can be thought of evidence, further more,the BPA of each classifier is determined, then the procession of the model identification must be improved.     

An improved HEED clustering algorithm for Wireless Sensor Network

In wireless sensor network, routing protocols which based on clustering have the advantages of energy consumption, topology management and data fusion. The HEED protocol, which generates cluster heads based on distributed algorithm, drives up the rate of clustering and creates well distributed cluster heads. However, it does not consider the mobility of nodes in the network. When the distance between neighbor nodes has changed, the AMRP method which decides the node belongs to different cluster heads would cause problems such as high energy consumption, short lifetime of network and so on. Responding to these problems, the paper proposes the S HEED, a clustering algorithm based on stability, which chooses the stability as a parameter of nodes when choosing a cluster head. With S HEED algorithm, the high energy consumption problem among cluster nodes and cluster heads caused by the mobility is tackled. The simulation experiment demonstrates that the S HEED algorithm lower the energy consumption of cluster heads and prolongs the network lifetime.     

The filtered back-projection algorithm of magnetic induction tomography with uniform magnetic excitation

Helmholtz coils produce uniform sinusoidal magnetic field in the center region, and the direction of magnetic field is approximate straight line, to help simplify the complexity of inverse problems. The simulation models of an 8-channel magnetic induction tomography measurement system are built, reconstructing conductivity distribution with filtered back projection algorithm. In the filtered back projection algorithm, the detected data is supplemented by the linear interpolation first, and then filtered by the Hamming filter, while adding a window filter to reduce the impact of around the coil from the divergence of the magnetic field. The different noise ratio of noise is added in the detected data to test noise suppression ability of the algorithm. The experiment results show that this filtered back projection algorithm can reconstruct the conductivity distribution under this model.     

Nondestructive Test of Surface Defect of Hot Slabs Based on Machine Vision at High Temperature State

Aimed at the problem of online detection of continuous slabs surface defect at high temperature, The CCD shutter control model was established using machine vision technology and equipment at high temperature states. The default clustering algorithm was provided. The CCD imaging equipment with dual cool modes of water and air was selected to ensure continuous work at high temperature conditions. The slab width of 3 000 mm can be detected, and the system can effectively restrain the influence of noise signals brought by oscillation marks on the surface of slabs and realize monitoring and clustering of the slab surface default.     

基于深度卷积神经网络的玉米病害识别

为了提高玉米病害的识别率,本文提出了一种在自然环境条件下基于深度卷积神经网络的玉米病害识别方法。该方法以玉米常见的10类病害为研究对象。算法模型是先将图像预处理,应用Triplet loss双卷积神经网络结构学习玉米图像特征,再使用SIFT算法提取图像纹理细节,最后通过Softmax对图像进行标签分类。训练集采用正常玉米图像与玉米病害图像相结合的方式,使用深度相似性网络学习正常玉米图像特征表示,再使用迁移学习方法学习玉米病害图像的特征,最后对特征进行分类识别。研究结果表明,该方法可准确识别10种常见玉米病害,正确率可达90%以上,为玉米病害的防治提供了有效的技术支持。     

多邻域保持嵌入的人脸识别方法

现有流形学习算法在学习人脸数据时,假设所有数据点位于单一低维嵌入流形之上,当数据点实际分布在不同的流形上时,单流形假设就会影响数据真实空间结构。为此提出一种基于多邻域保持嵌入（multiple neighborhood preserving embedding, M-NPE）的学习算法来发现不同类别数据在不同维度的低维嵌入空间中分布的多流形结构。首先,单独学习不同类别数据的流形,得到反映其本质特征的流形;再通过遗传算法搜索每个流形的最优维数;最后依据最小重构误差分类器对样本分类。在Extended Yale B和CMU PIE这2个大型人脸库上实验结果验证了该算法的有效性。     

Learning the BN Structure Based on the Sum of Mutual Information

The Crossing Entropy is defined to scale the similar level of two probability distribution. In many papers on learning BN structure,the Crossing Entropy was used as an indicator of measuring the learning accuracy of an algorithm.The known scoring metrics for learning BN structure is analyzed in this paper,then a new scoring metrics Sum of Mutual Information is proposed based on the information theory.At last,two algorithm for learning BN structure by SIM is represented.     

Application of Fuzzy-Set Clustering for Regional Typologies

Qualitative and quantitative procedures have been used to aggregate communities and counties for regional economic analysis. However, Once aggregated, communities and counties are perceived as homogeneous entities; this often belies the diversity that may exist. In order to capture the non-uniqueness of counties, fuzzy-set clustering procedures were employed to derive a typology of Nevada counties. Fuzzy-set clustering procedures employing fuzzy-set membership values and possibility theory derive county membership values associated for specific county clusters. Information from fuzzy partitions yields a means for posterior evaluation of county clusters which is independent of the algorithm producing them. From county membership values calculated from results of the fuzzy-set clustering analysis for Nevada, specific economic development programs for aggregate and individual counties are derived.     

Fuzzy c-means Cluster Over Gentic Algorithm

Fuzzy c-means (FCM) algorithm is dynamic cluster algorithm whose result often is local optimal decision. There often exists insignificant clustering in the result of Fuzzy c-means algorithm when traditional union, Intersection and inclusion work in fuzzy set. Our research indicates there are no insignificant clustering in the result of Fuzzy c-means algorithm over genetic algorithm and partial optimal solution can be avoided with this algorithm to a certain extent. The coding, select, corresponding crossover and mutation operators are designed. Finally we compared the performance of GFCM and FCM with testing data. Results show that the performance of GFCM is far better than FCM.     

多种分类器在农用地分等中的应用及其用法改良

以广东省第二次土壤普查成果资料为主要数据源,选取贝叶斯决策、BP神经网络、概率神经网络、聚类等分类方法分别对数据源进行分类;并且,笔者为了充分利用有监督学习分类准确率高和无监督学习无需标定的学习样本的优点,提出了基于监督--非监督的聚类算法,然后对上述五种方法的评价结果作了比较分析;实验表明文章提出的基于监督--非监督聚类方法只利用少量的有标定学习样本,即可得到较高的分类准确率,特别在少量样本时,该方法能得到比贝叶斯决策方法、BP神经网络和概率神经网络等监督学习方法更好的土地评价结果;在实际应用中,可以尝试结合监督和非监督学习的方法,实现分类正确率和获取大量有类标签的样本之间的折中。